llama_cpp 0.18.0 → 0.18.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 0a9263eee75a3d91907c711565799fd25820d8d2b0f0ae9818a24a0798b49bf4
-  data.tar.gz: 9f05051e8972baea44c4bd33f63190d5da8a62156c419fc02ec6b96ea3545c31
+  metadata.gz: a043b8beda10d1eed3dfd6ab21932f567fa9ff69a82e14048f19993bbcff9c3a
+  data.tar.gz: '08bba39155267edd1d5a99ad4170e44b3075fe552b096d16bac38d438a9004e5'
 SHA512:
-  metadata.gz: 2543f1022462d32694649f2226d097537672bd6921af4fe6948687e00ef007b2e0425110abc7a3240d5dceb25131a3641bc9614f0e1117f3a7a40dcc55b23190
-  data.tar.gz: 0c54ea7e7617e99f52b0f005e8f871d5f69423e92d350bc93562ba56b48ff85a88ee6c3f01e1e6a71a0782a1bcf212f4900c7b88229a63935349cbdabee95cfc
+  metadata.gz: 404c50dedbd90e8c457113145d3029f130a2c6cf8cd9aee9be867c03ac7061c271907649cfb59dd23ff1331d9c64eaf234434a2354c8f3079265a4efe7cd9ea0
+  data.tar.gz: b2b75ed8a826c5eb5683010dcf7ca0ad9f99d3f0cfd5c2016acf1717489e17eef005ad9eaa9997fb2529c9c75f40bc301c4c3804e24b3334036244dec69f4b92

data/CHANGELOG.md

@@ -1,6 +1,20 @@
+
+## [[0.18.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.18.1...v0.18.2)] - 2025-03-01
+
+- Change supported llama.cpp version to b4793
+  - Add `llama_model_n_head_kv` module function.
+  - Add `LLAMA_VOCAB_PRE_TYPE_GPT4O` constant.
+
+## [[0.18.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.18.0...v0.18.1)] - 2025-02-15
+
+- Change supported llama.cpp version to b4713
+  - Add `llama_sampler_init_top_n_sigma` module function.
+- Remove old type declaration file.
+
 ## [[0.18.0](https://github.com/yoshoku/llama_cpp.rb/compare/v0.17.10...v0.18.0)] - 2025-02-02
 
 **Breaking Changes**
+
 All the native extensions code was rewritten in C. The high-level API has been removed and replaced with a simple bindings library.
 The fast update speed of llama.cpp makes it difficult to keep up with the creation of this binding library.
 [As previously noted](https://github.com/yoshoku/llama_cpp.rb/blob/main/CHANGELOG.md#060---2023-09-30),

data/README.md

@@ -6,8 +6,6 @@
 
 llama_cpp.rb provides Ruby bindings for the [llama.cpp](https://github.com/ggerganov/llama.cpp).
 
-This gem is still under development and may undergo many changes in the future.
-
 ## Installation
 
 Install the llama.cpp. If you use homebrew, install it by executing:

data/ext/llama_cpp/llama_cpp.c

@@ -1452,6 +1452,20 @@ static VALUE rb_llama_model_n_head(VALUE self, VALUE model) {
   return INT2NUM(llama_model_n_head(model_wrapper->model));
 }
 
+/**
+ * @overload llama_model_n_head_kv(model)
+ *  @param [LlamaModel] model
+ *  @return [Integer]
+ */
+static VALUE rb_llama_model_n_head_kv(VALUE self, VALUE model) {
+  if (!rb_obj_is_kind_of(model, rb_cLlamaModel)) {
+    rb_raise(rb_eArgError, "model must be a LlamaModel");
+    return Qnil;
+  }
+  llama_model_wrapper* model_wrapper = get_llama_model_wrapper(model);
+  return INT2NUM(llama_model_n_head_kv(model_wrapper->model));
+}
+
 /**
  * @overload llama_model_rope_freq_scale_train(model)
  *  @param [LlamaModel] model
@@ -3367,6 +3381,20 @@ static VALUE rb_llama_sampler_init_xtc(VALUE self, VALUE p, VALUE t, VALUE min_k
   return TypedData_Wrap_Struct(rb_cLlamaSampler, &llama_sampler_data_type, sampler);
 }
 
+/**
+ * @overload llama_sampler_init_top_n_sigma(n)
+ *  @param [Float] n
+ *  @return [LlamaSampler]
+ */
+static VALUE rb_llama_sampler_init_top_n_sigma(VALUE self, VALUE n) {
+  if (!RB_FLOAT_TYPE_P(n)) {
+    rb_raise(rb_eArgError, "n must be a Float");
+    return Qnil;
+  }
+  struct llama_sampler* sampler = llama_sampler_init_top_n_sigma(NUM2DBL(n));
+  return TypedData_Wrap_Struct(rb_cLlamaSampler, &llama_sampler_data_type, sampler);
+}
+
 /**
  * @overload llama_sampler_init_mirostat(n_vocab, seed, tau, eta, m)
  *  @param [Integer] n_vocab
@@ -3881,6 +3909,7 @@ void Init_llama_cpp(void) {
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_CHAMELEON", INT2NUM(LLAMA_VOCAB_PRE_TYPE_CHAMELEON));
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_MINERVA", INT2NUM(LLAMA_VOCAB_PRE_TYPE_MINERVA));
   rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM", INT2NUM(LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM));
+  rb_define_const(rb_mLlamaCpp, "LLAMA_VOCAB_PRE_TYPE_GPT4O", INT2NUM(LLAMA_VOCAB_PRE_TYPE_GPT4O));
   /* llama_rope_type */
   /* Document-const: LlamaCpp::LLAMA_ROPE_TYPE_NONE */
   rb_define_const(rb_mLlamaCpp, "LLAMA_ROPE_TYPE_NONE", INT2NUM(LLAMA_ROPE_TYPE_NONE));
@@ -4700,6 +4729,9 @@ void Init_llama_cpp(void) {
   /* llama_model_n_head */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_n_head", rb_llama_model_n_head, 1);
 
+  /* llama_model_n_head_kv */
+  rb_define_module_function(rb_mLlamaCpp, "llama_model_n_head_kv", rb_llama_model_n_head_kv, 1);
+
   /* llama_model_rope_freq_scale_train */
   rb_define_module_function(rb_mLlamaCpp, "llama_model_rope_freq_scale_train", rb_llama_model_rope_freq_scale_train, 1);
 
@@ -5047,6 +5079,9 @@ void Init_llama_cpp(void) {
   /* llama_sampler_init_xtc */
   rb_define_module_function(rb_mLlamaCpp, "llama_sampler_init_xtc", rb_llama_sampler_init_xtc, 4);
 
+  /* llama_sampler_init_top_n_sigma */
+  rb_define_module_function(rb_mLlamaCpp, "llama_sampler_init_top_n_sigma", rb_llama_sampler_init_top_n_sigma, 1);
+
   /* llama_sampler_init_mirostat */
   rb_define_module_function(rb_mLlamaCpp, "llama_sampler_init_mirostat", rb_llama_sampler_init_mirostat, 5);
 

data/lib/llama_cpp/version.rb

@@ -3,8 +3,8 @@
 # llama_cpp.rb provides Ruby bindings for the llama.cpp.
 module LlamaCpp
   # The version of llama_cpp.rb you install.
-  VERSION = '0.18.0'
+  VERSION = '0.18.2'
 
   # The supported version of llama.cpp.
-  LLAMA_CPP_VERSION = 'b4611'
+  LLAMA_CPP_VERSION = 'b4793'
 end

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.18.0
+  version: 0.18.2
 platform: ruby
 authors:
 - yoshoku
 bindir: exe
 cert_chain: []
-date: 2025-02-02 00:00:00.000000000 Z
+date: 2025-03-01 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email:
@@ -26,7 +26,6 @@ files:
 - ext/llama_cpp/llama_cpp.h
 - lib/llama_cpp.rb
 - lib/llama_cpp/version.rb
-- sig/llama_cpp.rbs
 homepage: https://github.com/yoshoku/llama_cpp.rb
 licenses:
 - MIT

data/sig/llama_cpp.rbs

@@ -1,366 +0,0 @@
-module LLaMACpp
-  VERSION: String
-  LLAMA_CPP_VERSION: String
-  LLAMA_DEFALUT_SEED: String
-
-  LLAMA_FILE_MAGIC_GGLA: String
-  LLAMA_FILE_MAGIC_GGSN: String
-  LLAMA_FILE_MAGIC_GGSQ: String
-  LLAMA_SESSION_MAGIC: String
-  LLAMA_SESSION_VERSION: String
-  LLAMA_STATE_SEQ_MAGIC: String
-  LLAMA_STATE_SEQ_VERSION: String
-
-  LLAMA_VOCAB_TYPE_NONE: Integer
-  LLAMA_VOCAB_TYPE_SPM: Integer
-  LLAMA_VOCAB_TYPE_BPE: Integer
-  LLAMA_VOCAB_TYPE_WPM: Integer
-  LLAMA_VOCAB_TYPE_UGM: Integer
-  LLAMA_VOCAB_TYPE_RWKV: Integer
-
-  LLAMA_VOCAB_PRE_TYPE_DEFAULT: Integer
-  LLAMA_VOCAB_PRE_TYPE_LLAMA3: Integer
-  LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM: Integer
-  LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER: Integer
-  LLAMA_VOCAB_PRE_TYPE_FALCON: Integer
-  LLAMA_VOCAB_PRE_TYPE_MPT: Integer
-  LLAMA_VOCAB_PRE_TYPE_STARCODER: Integer
-  LLAMA_VOCAB_PRE_TYPE_GPT2: Integer
-  LLAMA_VOCAB_PRE_TYPE_REFACT: Integer
-  LLAMA_VOCAB_PRE_TYPE_COMMAND_R: Integer
-  LLAMA_VOCAB_PRE_TYPE_STABLELM2: Integer
-  LLAMA_VOCAB_PRE_TYPE_QWEN2: Integer
-  LLAMA_VOCAB_PRE_TYPE_OLMO: Integer
-  LLAMA_VOCAB_PRE_TYPE_DBRX: Integer
-  LLAMA_VOCAB_PRE_TYPE_SMAUG: Integer
-  LLAMA_VOCAB_PRE_TYPE_PORO: Integer
-  LLAMA_VOCAB_PRE_TYPE_CHATGLM3: Integer
-  LLAMA_VOCAB_PRE_TYPE_CHATGLM4: Integer
-  LLAMA_VOCAB_PRE_TYPE_VIKING: Integer
-  LLAMA_VOCAB_PRE_TYPE_JAIS: Integer
-  LLAMA_VOCAB_PRE_TYPE_TEKKEN: Integer
-  LLAMA_VOCAB_PRE_TYPE_SMOLLM: Integer
-  LLAMA_VOCAB_PRE_TYPE_CODESHELL: Integer
-  LLAMA_VOCAB_PRE_TYPE_BLOOM: Integer
-  LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH: Integer
-  LLAMA_VOCAB_PRE_TYPE_EXAONE: Integer
-
-  LLAMA_TOKEN_ATTR_UNDEFINED: Integer
-  LLAMA_TOKEN_ATTR_UNKNOWN: Integer
-  LLAMA_TOKEN_ATTR_UNUSED: Integer
-  LLAMA_TOKEN_ATTR_NORMAL: Integer
-  LLAMA_TOKEN_ATTR_CONTROL: Integer
-  LLAMA_TOKEN_ATTR_USER_DEFINED: Integer
-  LLAMA_TOKEN_ATTR_BYTE: Integer
-  LLAMA_TOKEN_ATTR_NORMALIZED: Integer
-  LLAMA_TOKEN_ATTR_LSTRIP: Integer
-  LLAMA_TOKEN_ATTR_RSTRIP: Integer
-  LLAMA_TOKEN_ATTR_SINGLE_WORD: Integer
-
-  LLAMA_FTYPE_ALL_F32: Integer
-  LLAMA_FTYPE_MOSTLY_F16: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_0: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_1: Integer
-  LLAMA_FTYPE_MOSTLY_Q8_0: Integer
-  LLAMA_FTYPE_MOSTLY_Q5_0: Integer
-  LLAMA_FTYPE_MOSTLY_Q5_1: Integer
-  LLAMA_FTYPE_MOSTLY_Q2_K: Integer
-  LLAMA_FTYPE_MOSTLY_Q3_K_S: Integer
-  LLAMA_FTYPE_MOSTLY_Q3_K_M: Integer
-  LLAMA_FTYPE_MOSTLY_Q3_K_L: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_K_S: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_K_M: Integer
-  LLAMA_FTYPE_MOSTLY_Q5_K_S: Integer
-  LLAMA_FTYPE_MOSTLY_Q5_K_M: Integer
-  LLAMA_FTYPE_MOSTLY_Q6_K: Integer
-  LLAMA_FTYPE_MOSTLY_IQ2_XXS: Integer
-  LLAMA_FTYPE_MOSTLY_IQ2_XS: Integer
-  LLAMA_FTYPE_MOSTLY_Q2_K_S: Integer
-  LLAMA_FTYPE_MOSTLY_IQ3_XS: Integer
-  LLAMA_FTYPE_MOSTLY_IQ3_XXS: Integer
-  LLAMA_FTYPE_MOSTLY_IQ1_S: Integer
-  LLAMA_FTYPE_MOSTLY_IQ4_NL: Integer
-  LLAMA_FTYPE_MOSTLY_IQ3_S: Integer
-  LLAMA_FTYPE_MOSTLY_IQ3_M: Integer
-  LLAMA_FTYPE_MOSTLY_IQ4_XS: Integer
-  LLAMA_FTYPE_MOSTLY_IQ1_M: Integer
-  LLAMA_FTYPE_MOSTLY_BF16: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_0_4_4: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_0_4_8: Integer
-  LLAMA_FTYPE_MOSTLY_Q4_0_8_8: Integer
-  LLAMA_FTYPE_MOSTLY_TQ1_0: Integer
-  LLAMA_FTYPE_MOSTLY_TQ2_0: Integer
-  LLAMA_FTYPE_GUESSED: Integer
-
-  LLAMA_KV_OVERRIDE_TYPE_INT: Integer
-  LLAMA_KV_OVERRIDE_TYPE_FLOAT: Integer
-  LLAMA_KV_OVERRIDE_TYPE_BOOL: Integer
-  LLAMA_KV_OVERRIDE_TYPE_STR: Integer
-
-  LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED: Integer
-  LLAMA_ROPE_SCALING_TYPE_NONE: Integer
-  LLAMA_ROPE_SCALING_TYPE_LINEAR: Integer
-  LLAMA_ROPE_SCALING_TYPE_YARN: Integer
-  LLAMA_ROPE_SCALING_TYPE_MAX_VALUE: Integer
-
-  LLAMA_POOLING_TYPE_UNSPECIFIED: Integer
-  LLAMA_POOLING_TYPE_NONE: Integer
-  LLAMA_POOLING_TYPE_MEAN: Integer
-  LLAMA_POOLING_TYPE_CLS: Integer
-  LLAMA_POOLING_TYPE_LAST: Integer
-
-  LLAMA_ATTENTION_TYPE_UNSPECIFIED: Integer
-  LLAMA_ATTENTION_TYPE_CAUSAL: Integer
-  LLAMA_ATTENTION_TYPE_NON_CAUSAL: Integer
-
-  LLAMA_SPLIT_MODE_NONE: Integer
-  LLAMA_SPLIT_MODE_LAYER: Integer
-  LLAMA_SPLIT_MODE_ROW: Integer
-
-  def self?.backend_init: () -> void
-  def self?.backend_free: () -> void
-  def self?.numa_init: (Integer) -> void
-  def self?.model_quantize: (input_path: String, output_path: String, params: ModelQuantizeParams) -> void
-  def self?.generate: (::LLaMACpp::Context, String,
-    ?n_predict: Integer, ?n_threads: Integer, ?n_keep: Integer, ?n_batch: Integer,
-    ?repeat_last_n: Integer, ?repeat_penalty: Float, ?frequency: Float, ?presence: Float,
-    ?top_k: Integer, ?top_p: Float, ?tfs_z: Float, ?typical_p: Float, ?temperature: Float) -> String
-  def self?.print_system_info: () -> void
-  def self?.time_us: () -> Integer
-  def self?.max_devices: () -> Integer
-  def self?.supports_mmap?: () -> bool
-  def self?.supports_mlock?: () -> bool
-  def self?.supports_gpu_offload?: () -> bool
-
-  class TokenData
-    public
-
-    def initialize: (id: Integer, logit: Float, p: Float) -> void
-    def id: () -> Integer
-    def id=: (Integer) -> Integer
-    def logit: () -> Float
-    def logit=: (Float) -> Float
-    def p: () -> Float
-    def p=: (Float) -> Float
-  end
-
-  class TokenDataArray
-    public
-
-    def initialize: (Array[::LLaMACpp::TokenData], ?sorted: bool) -> void
-    def size: () -> Integer
-    def sorted: () -> bool
-  end
-
-  class Model
-    public
-
-    def initialize: (model_path: String, params: ::LLaMACpp::ModelParams) -> void
-                  | () -> void
-    def empty?: () -> bool
-    def free: () -> void
-    def load: (model_path: String, params: ::LLaMACpp::ModelParams) -> void
-    def vocab_type: () -> Integer
-    def rope_type: () -> Integer
-    def n_vocab: () -> Integer
-    def n_ctx_train: () -> Integer
-    def n_embd: () -> Integer
-    def n_layer: () -> Integer
-    def rope_freq_scale_train: () -> Float
-    def token_to_piece: (Integer, ?lstrip: Integer, ?special: bool) -> String
-    def tokenize: (text: String, ?n_max_tokens: Integer, ?add_bos: bool, ?special: bool) -> Array[Integer]
-    def desc: () -> String
-    def size: () -> Integer
-    def n_params: () -> Integer
-    def text: (Integer) -> String
-    def score: (Integer) -> Float
-    def token_attr: (Integer) -> Integer
-    def token_bos: () -> Integer
-    def token_eos: () -> Integer
-    def token_cls: () -> Integer
-    def token_sep: () -> Integer
-    def token_nl: () -> Integer
-    def token_pad: () -> Integer
-    def add_bos_token?: () -> bool
-    def add_eos_token?: () -> bool
-    def token_prefix: () -> Integer
-    def token_middle: () -> Integer
-    def token_suffix: () -> Integer
-    def token_eot: () -> Integer
-    def token_is_eog?: (Integer) -> bool
-    def token_is_control?: (Integer) -> bool
-    def has_encoder?: () -> bool
-    def has_decoder?: () -> bool
-    def decoder_start_token: () -> Integer
-    def is_recurrent?: () -> bool
-    def detokenize: (Array[Integer], ?remove_special: bool, ?unparse_special: bool) -> String
-  end
-
-  class ModelKVOverride
-    public
-
-    def key: () -> String
-    def tag: () -> Integer
-    def val_i64: () -> Integer
-    def val_f64: () -> Float
-    def val_bool: () -> bool
-    def val_str: () -> String
-  end
-
-  class ModelParams
-    public
-
-    def n_gpu_layers: () -> Integer
-    def n_gpu_layers=: (Integer) -> Integer
-    def split_mode: () -> Integer
-    def split_mode=: (Integer) -> Integer
-    def main_gpu: () -> Integer
-    def main_gpu=: (Integer) -> Integer
-    def tensor_split: () -> Array[Float]
-    def vocab_only: () -> bool
-    def vocab_only=: (bool) -> bool
-    def use_mmap: () -> bool
-    def use_mmap=: (bool) -> bool
-    def use_mlock: () -> bool
-    def use_mlock=: (bool) -> bool
-    def check_tensors: () -> bool
-    def check_tensors=: (bool) -> bool
-  end
-
-  class Batch
-    public
-
-    def self.get_one: (tokens: Array[Integer], n_tokens: Integer, pos_zero: Integer, seq_id: Integer) -> ::LLaMACpp::Batch
-    def initialize: (max_n_token: Integer, n_embd: Integer, max_n_seq: Integer) -> void
-    def n_tokens=: (Integer) -> Integer
-    def n_tokens: () -> Integer
-    def all_pos_zero=: (Integer) -> Integer
-    def all_pos_zero: () -> Integer
-    def all_pos_one=: (Integer) -> Integer
-    def all_pos_one: () -> Integer
-    def all_seq_id=: (Integer) -> Integer
-    def all_seq_id: () -> Integer
-    def set_token: (Integer, Integer) -> Integer
-    def get_token: (Integer) -> Integer
-    def set_pos: (Integer, Integer) -> Integer
-    def get_pos: (Integer) -> Integer
-    def set_n_seq_id: (Integer, Integer) -> Integer
-    def get_n_seq_id: (Integer) -> Integer
-    def set_seq_id: (Integer, Integer, Integer) -> Integer
-    def get_seq_id: (Integer, Integer) -> Integer
-    def set_logit: (Integer, bool) -> bool
-    def get_logit: (Integer) -> bool
-  end
-
-  class Context
-    public
-
-    attr_reader model: ::LLaMACpp::Model
-
-    def initialize: (model: ::LLaMACpp::Model, params: ::LLaMACpp::ContextParams) -> void
-    def embeddings: () -> Array[Float]
-    def embeddings_ith: (Integer) -> Array[Float]
-    def embeddings_seq: (Integer) -> Array[Float]
-    def encode: (::LLaMACpp::Batch) -> void
-    def decode: (::LLaMACpp::Batch) -> void
-    def logits: () -> Array[Float]
-    def set_embeddings: (bool) -> void
-    def set_n_threads: (n_threads: Integer, n_threads_batch: Integer) -> void
-    def n_ctx: () -> Integer
-    def n_batch: () -> Integer
-    def n_ubatch: () -> Integer
-    def n_seq_max: () -> Integer
-    def n_threads: () -> Integer
-    def n_threads_batch: () -> Integer
-    def kv_cache_token_count: () -> Integer
-    def kv_cache_clear: () -> void
-    def kv_cache_seq_rm: (Integer, Integer,Integer) -> void
-    def kv_cache_seq_cp: (Integer, Integer,Integer, Integer) -> void
-    def kv_cache_seq_keep: (Integer) -> void
-    def kv_cache_seq_add: (Integer, Integer, Integer, Integer) -> void
-    def kv_cache_seq_div: (Integer, Integer, Integer, Integer) -> void
-    def kv_cache_seq_pos_max: (Integer) -> Integer
-    def kv_cache_defrag: () -> void
-    def kv_cache_update: () -> void
-    def set_causal_attn: (bool) -> void
-    def synchronize: () -> void
-    def load_session_file: (session_path: String) -> void
-    def save_session_file: (session_path: String, session_tokens: Array[Integer]) -> void
-    def apply_control_vector: (data: Array[Float], n_embd: Integer, il_start: Integer, il_end: Integer) -> void
-    def pooling_type: () -> Integer
-  end
-
-  class ContextParams
-    public
-
-    def n_ctx: () -> Integer
-    def n_ctx=: (Integer) -> Integer
-    def n_batch: () -> Integer
-    def n_batch=: (Integer) -> Integer
-    def n_ubatch: () -> Integer
-    def n_ubatch=: (Integer) -> Integer
-    def n_seq_max: () -> Integer
-    def n_seq_max=: (Integer) -> Integer
-    def n_threads: () -> Integer
-    def n_threads=: (Integer) -> Integer
-    def n_threads_batch: () -> Integer
-    def n_threads_batch=: (Integer) -> Integer
-    def rope_scaling_type=: (Integer) -> Integer
-    def rope_scaling_type: () -> Integer
-    def pooling_type=: (Integer) -> Integer
-    def pooling_type: () -> Integer
-    def attention_type=: (Integer) -> Integer
-    def attention_type: () -> Integer
-    def rope_freq_base=: (Float) -> Float
-    def rope_freq_base: () -> Float
-    def rope_freq_scale=: (Float) -> Float
-    def rope_freq_scale: () -> Float
-    def yarn_ext_factor=: (Float) -> Float
-    def yarn_ext_factor: () -> Float
-    def yarn_attn_factor=: (Float) -> Float
-    def yarn_attn_factor: () -> Float
-    def yarn_beta_fast=: (Float) -> Float
-    def yarn_beta_fast: () -> Float
-    def yarn_beta_slow=: (Float) -> Float
-    def yarn_beta_slow: () -> Float
-    def yarn_orig_ctx=: (Integer) -> Integer
-    def yarn_orig_ctx: () -> Integer
-    def defrag_thold=: (Float) -> Float
-    def defrag_thold: () -> Float
-    def type_k=: (Integer) -> Integer
-    def type_k: () -> Integer
-    def type_v=: (Integer) -> Integer
-    def type_v: () -> Integer
-    def logits_all: () -> bool
-    def logits_all=: (bool) -> bool
-    def embeddings: () -> bool
-    def embeddings=: (bool) -> bool
-    def offload_kqv: () -> bool
-    def offload_kqv=: (bool) -> bool
-    def flash_attn: () -> bool
-    def flash_attn=: (bool) -> bool
-  end
-
-  class ModelQuantizeParams
-    public
-
-    attr_reader params: ::LLaMACpp::ModelParams
-
-    def n_thread: () -> Integer
-    def n_thread=: (Integer) -> Integer
-    def ftype: () -> Integer
-    def ftype=: (Integer) -> Integer
-    def allow_quantization: () -> bool
-    def allow_quantization=: (bool) -> bool
-    def quantize_output_tensor: () -> bool
-    def quantize_output_tensor=: (bool) -> bool
-    def only_copy: () -> bool
-    def only_copy=: (bool) -> bool
-    def pure: () -> bool
-    def pure=: (bool) -> bool
-    def keep_split: () -> bool
-    def keep_split=: (bool) -> bool
-  end
-
-  class Params = ContextParams
-end